We Are Looking at AI Harm From the Wrong End
中文摘要
AI的危害并非始于幻觉，而是在开发阶段就已经存在。我们必须关注模型到达用户之前所产生的风险。
English Summary
AI harms originate during the development phase, long before hallucinations occur. We must re-examine risks that emerge before models reach users.
原文节选
The dangers of language models do not begin with hallucinations. Many of them appear before a model reaches its first user. Continue reading on Medium »